starlark: bring floating-point into spec compliance (#313)

This change makes go.starlark.net's floating-point implementation
match the proposed spec (see https://github.com/bazelbuild/starlark/pull/119),
and thus much more closely match the behavior of the Java implementation.

The major changes are:
- Float values are totally ordered; NaN compares greater than +Inf.
- The string form of a finite float value always contains an exponent
  or a decimal point, so they are self-evidently not int values.
- Operations that would cause a large integer to become rounded to
  an infinite float are now an error.

The resolve.AllowFloat boolean, and the corresponding -float command-line
flag, now have no effect. Floating point support is always enabled.

1 files changed, 0 insertions, 1 deletions

diff --git a/starlarkstruct/struct_test.go b/starlarkstruct/struct_test.go
index 8e6a93d..5de7020 100644
--- a/starlarkstruct/struct_test.go
+++ b/starlarkstruct/struct_test.go
@@ -19,7 +19,6 @@ func init() {
 	// The tests make extensive use of these not-yet-standard features.
 	resolve.AllowLambda = true
 	resolve.AllowNestedDef = true
-	resolve.AllowFloat = true
 	resolve.AllowSet = true
 }